Notes: Thomson's "Neocortical Layer 6, A Review"

[mrcslws]

I read Neocortical Layer 6, A Review. It took a while.

Here are my notes. I’ve tried to make them useful to others.

We haven’t discussed these notes at Numenta yet. All errors here are my own. I think it’s mostly right.

The paper focused on L6 in primary sensory cortex, e.g. V1. There may be differences in L6 of higher regions.

I also used some material from Zarrinpar, Callaway - Local connections to specific types of layer 6 neurons in the rat visual cortex. (2006). I haven’t finished reading that paper, but it seems good.

A network diagram

This image focuses on interactions with the thalamus. It doesn’t picture the long horizontal connections between L5 and L6.

Note the number of pathways up the hierarchy. For example, V1 sends information “up” to V2 via:

• Connections from V1 L5 to V2 L4
• V1 L5 → the thalamus (proximal)
• V1 Lower L6 → the thalamus (proximal)

It’s interesting that the thalamus doesn’t project to L6 in the “Secondary Sensory Cortex”, e.g. V2.

Inputs to L6

Primary sensory thalamus provides proximal input to some/all L6 cells.

Secondary / association thalamus does not provide input to L6. So if L6 cells need access to this, they’ll have to reach dendrites up into L4 / L5.

Very long horizontally oriented axons project between “deep layer” pyramids. So there’s plenty of opportunity for cross-column and inter-areal connections within the deep layers.

Here’s a some partly compatible, partly incompatible data from Page 4:

In a retrograde labelling study, the most profuse, longer distance, cortical inputs to layer 6 of the whisker barrel field came from motor cortex, with sparser inputs provided by other cortical somatosensory regions, thalamic afferents (from VPm, Pom, and the intralaminar nuclei) and the claustrum.

In other words, in this study, primary sensory L6 received lots of input from:

• Other parts of the cortex, especially the motor cortex
• Primary sensory thalamus for whiskers
• Secondary sensory thalamus for whiskers

This last one confused me. Elsewhere this paper says that secondary sensory thalamus only projects to L1-5 in primary sensory cortex.

The cells

Layer 6 seems to have a few things going on.

Cortico-thalamic (CT) pyramidal cells

30-50% of the cells in L6 are CT cells. They project their axons to the thalamus, by definition.

CT cells provide distal input to the primary sensory thalamus. For example, CT cells in V1 provide modulatory input to the LGN. So the LGN gets proximal input from the retina and distal input from L6 of the region that it projects into.

These connections are reciprocal. CT cells will selectively output to the region of primary sensory thalamus from which they receive direct input.

L6 CT cells receive a lot of inhibitory input via proximal synapses.

CT cells have two distinct types. They don’t really have names. The paper uses a couple naming schemes:

• “Primary sensory” and “Non-primary” – This naming is deceptive. All CT cells project to the primary sensory thalamus.
• “Specific” and “Non-specific” – This naming is more accurate, but it’s only used in a single caption of this paper (Figure 1). Usually these terms are used to describe thalamus, not cortex.

The first type is found in Upper L6. The second type is found in Lower L6. Just kidding, this isn’t always true. In another paper the “Lower L6 CT” cells were indeed found only in the lower third of L6, while the “Upper L6 CT” cells were found throughout L6.

Regardless, here let’s call them “Upper L6 CT” and “Lower L6 CT” cells.

CT cells: Upper Layer 6

These CT cells project axons to:

• Layer 4
• Primary sensory thalamus
• Thalamic reticular nucleus (nRT)

Zarrinpar and Callaway characterize the dendrites:

characterized by an apical dendrite with perpendicular branches in layer 5B and with one dendritic tuft that extends through 2 layers (layer 5A and layer 4)

This study found that Upper L6 CT cells receive excitatory input from layers 4, 5B, and 6. This may vary by species. This input may be arriving through axons projected into L6, or through L6’s distal dendrites in L4/5.

Upper L6 → L4

There are a lot of these connections. These connections don’t seem to cause many spikes in L4. It’s unclear whether this means:

• Upper L6 is connecting primarily to inhibitory interneurons in L4,
• …or Upper L6 is providing distal “modulatory” input to L4 cells,
• …or both.

Dendrites in L5 and L4

L6 projects distal dendrites up through L5 and L4. This might be a way of getting input from these cells. It also might be a way of getting “feedback” input from higher levels of thalamus.

L4 and thalamic targets are correlated

Here’s an interesting fact: A Upper L6 CT cell modulates the same region of thalamus that provides input to the L4 area that this L6 CT cell is also modulating.

This is like the reciprocal connections between CT cells and the thalamus, but going one step further and adding L4 into the mix.

For example:

• Some L6 cells project axons to parvocellular LGN regions in the thalamus. These cells also project to lower 4c, which receives parvocellular input.
• This is also seen with magnocellular LGN and upper layer 4.

From the paper:

These findings support an earlier proposal, that layer 6 pyramidal neurones target both specific sub-populations of LGN relay cells and the specific cells in layer 4 that are postsynaptic to those LGN cells (Lund and Boothe, 1975).

These correlated connections are applicable to combinations of:

• L6 CT axons to the thalamus
• L6 CT axons to L4
• L6 CT dendrites to L4
• Thalamic axons to L6 CT cells
• Thalamic axons to L4

CT cells: Lower Layer 6

These cells seem to be always located in Lower Layer 6.

Lower L6 CT cells project their axons to:

• Primary sensory thalamus - distal
• Secondary / associative thalamus - proximal. This thalamus also receives proximal input from L5 cells in primary sensory cortex.
• L5, L6
• Occasionally L2/3

Zarrinpar and Callaway characterize the dendrites:

characterized by an apical dendrite that had many perpendicular branches in layer 6, sparse perpendicular branches in layer 5B, and a tuft of dendritic branches in layer 5A, with none of the dendrites projecting into layer 4

This study found that Lower L6 CT cells receive most of their excitatory input from layer 6, with a minor component from layer 5B.

How consistent is this?

The paper implies that every L6 cell that projects to L4 also projects to the primary sensory thalamus. But there’s at least one exception in the paper. Page 9:

Mid layer 4C the zone of M and P combination, received from the middle of layer 6 (which did not project to the LGN), …

I don’t know where else these mid-L6 CT cells project, or whether they are “CT” cells at all.

Unique to L6 CT: “Facilitating” connections

Usually, pyramidal cells have “depressing” synapses. The initial activation of a pyramidal cell causes its output synapses to quickly excite the target dendrites. Subsequent activations have a lesser effect, since the synapse essentially needs to recharge. The name “depressing” focuses on the fact that subsequent activations have a lesser effect.

L6 CT cells have “facilitating” output. Subsequent activations of the cell have a greater effect on its outgoing synapses. Reaching the spike threshold often requires multiple firings of the synapse, so CT cells are slower at causing spikes in the postsynaptic cell.

This is true of all observed L6 CT axons, including those to L4 and to the thalamus.

Additionally, CC and CT cells themselves have different firing patterns. CC cells (and claustrum-projecting cells) will fire one to three short spikes at the start of a depolarizing pulse, then they’ll stop spiking despite additional current. CT cells don’t stop firing while the cell is above firing threshold. So the cell firing pattern seems to match the synapse behavior.

In this case, I think the terms “facilitating” and “depressing” are a strange way of describing this phenomenon. The only parameter being changed for the synapse is its release probability ‘p’. L6 CT synapses have a lower release probability, so it takes more presynaptic spikes to cause a postsynaptic spike. So there’s essentially a time delay and maybe a higher spike threshold for cells receiving L6 CT input. All this facilitation-depression talk might just be a roundabout way of saying that. But keep in mind that I’m inexperienced.

Corticocortical (CC) pyramidal cells

CC cells send long horizontal axons which form connections across cortical columns and cortical areas, e.g. somatosensory and motor.

Rarely innervate interneurons. Preferentially innervate pyramids.

These connections are “depressing”, i.e. they are faster than those of CT cells.

Major excitatory inputs come from L5 and L6.

These cells can be further subdivided into: “bipolar”, “inverted pyramid”, and “non-tufted putative upright” cells.

Bipolar cells

characterized by 2 large-diameter, vertically oriented spiny dendrites. One large-diameter dendrite projects toward the pia, whereas the other projects into the white matter.

Receive input from Bipolar from 5a, 5b, 6, and a very small input from L4

Inverted pyramids

characterized by pyramidal soma but a thick large-diameter dendrite (arrow) that did not project from the pial side of the soma.

Receive input near exclusively from L6.

Non-tufted pyramid

a.k.a “Non-tufted putative upright CC cells”

characterized by long angular branches from the apical den- drite, especially at the layer 5B/6 border. Apical dendrite does not end in a tuft.

Inputs:

• Mostly L5/L6
• Minor input from L2/3
• Very small input from L4

Claustrum projecting pyramidal cell

Axons go to:

• the claustrum
• L5 and L6. Long, horizontal axons.

These connections are “depressing”, i.e. they are faster than those of CT cells.

Apical dendrite reaches up into to Layer 1. There’s “at best a meager” apical tuft in L1.

Rarely innervate interneurons. Preferentially innervate pyramids.

Interneurons

Edit: Click the image to see it. I’m sorry, I think this is a bug in Discourse.

Receive most of the excitatory input from within L6. Some comes from L2/L3. Most input from L2/L3 goes into these multipolar interneurons.

Remember, when we talk about inhibitory input from L4 to L6, we’re talking about cells that live in L4. Here we’re talking about cells that live in L6.

Inhibitory inputs and outputs

Layer 4 projects lots of inhibitory axons to Layer 6. This output comes from multiple interneuron types:

• Fast-spiking basket cells
• Slower-spiking double bouquet cells

Similarly, Layer 6 projects inhibitory connections to Layer 4, and also to other layers:

• Fast-spiking basket cells L6 → L4
• Martinotti cells target distal dendrites L6 → all layers 1 through 6.

This relationship between L4 and L6 is similar to the relationship between L3 and L5. In each case, maybe the two layers are in some sense “kept in sync” by the basket cells sending axons in both directions.

Additionally, Upper L6 CT cells often connect to inhibitory interneurons in L4, so there’s yet another way L6 can inhibit L4 cells. These connections are “facilitating”, so they’re slower. The paper suggests it’s possible these L4 interneurons initially respond to feedforward thalamic input, and then their response is prolonged by these L6 connections, but this is just speculation.

On the other hand, L2/L3 projects to some L6 interneurons, and L3 projects to L4 and L5 interneurons. Inhibitory connections are all over the place, so maybe these L3<->L5 and L4<->L6 similarities are imaginary.

L6 CT cells prefer outputting to inhibitory interneurons in L6. The paper points out that selective innervation like this is unusual within a layer, but it happens elsewhere between layers. L3 cells prefer connecting to inhibitory interneurons in L4 and L5.

Interconnections within L6

The paper says:

CC pyramids innervate other pyramids much more frequently (>4×) than CT cells do.

But it should be noted that this is measuring the number of release sites, not the number of connections. This number is weighted by the strength of each connection.

The connections made by CC cell axons with their near neighbours involve larger numbers of release sites … The outputs of CT cells onto other pyramids typically utilise fewer release sites …

Another paper that actually looked at the connections gives a lower estimate, though the CC cells still win:

Pyramidal cells with cortico-cortical like morphology were 2–4 times more likely to innervate other pyramidal cells than were cortico-thalamic like cells, but less likely to innervate inhibitory interneurons.

Also keep in mind that CT cells are 30-50% of the pyramidal cells in L6, so they’ll still be a significant portion of the input to other pyramidal cells.

The Thomson paper continues to say:

[Layer 6 CC cells] very rarely innervate layer 6 interneurones, however. The opposite is true for layer 6 CT cells which rarely innervate other layer 6 pyramids, but frequently innervate inhibitory interneurones.

But again this might be overstated. In the source paper, they surveyed 59 interneuron + pyramid pairs. In 12 of these pairs, the pyramid provided input to the interneuron. 10 of these 12 pyramids were inspected: 8 of them were CT, 2 of them were CC. They did not check how many of these 59 pyramids were CC and how many were CT. So this doesn’t say anything definitive.

Long receptive fields

In V1, L6 cells have long receptive fields. This happens via cross-column connections to L5.

When layer 5 was blocked very locally with GABA application, layer 6 cells lost the component of their receptive fields that cor- responded in visual space with the inactivated region of layer 5.

We don’t know which pyramidal cells have these long receptive fields. It might be the CC cells, the CT cells, or both. Similarly, we don’t know which cells receive this cross-column input from L5.

Approximately 17% of layer 6 cells in V1 had receptive fields greater than 8 degrees. The mean receptive field length in L6 is 2.8 degrees. 18% are 1 degree or less.

The maximum receptive field length in L4 has been reported to be 4 degrees.

Simple cells vs complex cells

Here’s an angle that may be useful: note the typical cell type in each layer.

• Simple: 4, Upper L6
• Complex: 2, 3, 5, Lower L6

Layer 6: Three parts?

It might make sense to split L6 into thirds.

The Zarrinpar, Callaway paper found that our “Lower L6” cells only occur in the lower third of the layer.

As seen above in “L4 and thalamic targets are correlated”, in the LGN → L6 pathway, there are particular connections for upper, middle, and lower L6.

[alavin]

@mrcslws led a journal club meeting today to go over this paper. Here are some notes from our discussion:

• L6 has ~2x the number of cells as any other layer in the same region of cortex (rat somatosensory, from HBP database).
• Network diagram
  • L6 definitely receives inputs from the thalamus, which is not reflected in the network diagram
  • L5 outputs either go through the thalamus and up to the next region, or goes subcortical. Here she shows a direct path from L5 into L4 of the next region.
  • Shown here is L6b connects broadly to L1 of the region below, although she doesn’t discuss this anywhere in the paper.
• Overall this paper gives us a clear distinction between 6a and 6b.
• “Simple” and “complex”are used wrt the classical neuroscience definitions—simple/complex receptive field properties, like in V1.
  • E.g. simple can respond to a very specific input, while complex will respond to a variety.
  • We can think of simple cells as doing spatial pooling
  • Tread lightly, simple vs. complex labels can be misleading
• L6 receiving a lot of input from motor cortex—i.e., the where pathway.
• CT cells make up 30-50% of the population, but we’re unsure what the claustrum is doing.
• L4c is (typically) only found in primate V1.
• L6 cells project long distances, and make connections locally —> implies they’re transmitting something between what and where pathways.
• Shouldn’t glean much from the cell reconstruction diagrams; methods of doing so are poor, and just b/c there are synapses doesn’t mean there’s activity.

[s.aleksashenko]

Quick note about the claustrum - I’m almost certain that it handles consciousness (consciousness “lives” there). There is little research on it because of practical reasons, so the evidence is circumstantial, but here it is:

1. Putting an electrode into the claustrum allows for an “on-off” switch for consciousness http://www.ncbi.nlm.nih.gov/pubmed/24967698
2. Salvia divinorum is an agonist of k-opiod receptors that are most concentrated in the claustrum http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3935397/
   Why is this important? Users of the drug describe the experience as being conscious, but not having access to the sensory stream or memories (don’t have an academic source for this one, just lots of hippie friends, also in the linked paper there is an “exploratory analysis of trip reports”, aka hippie stories). My guess is that the drug blocks input from the cortex to the claustrum.

Don’t want to get into philosophy, but since the term is largely debated, in my view consciousness can be thought of as the pinnacle of attention.

P.S. I’m assuming that anyone thinking about the claustrum has read What is the function of the claustrum? by Crick and Koch, if not - here’s a link

[mrcslws]

Here’s some followup about the L6b → Thalamus “up the hierarchy” proximal input.

Mentions in this paper

Here are a few places this is described in the paper.

Beginning of Page 9:

These cortical projections to secondary or association areas often form larger boutons and complex glomerular synapses with the proximal dendrites of relay cells (Usrey and Fitzpatrick, 1996; Murphy et al., 2000)

Middle of Page 9:

The output synapses from V2 CT cells to the pulvinar complex, a large region of association thalamus, occupy proximal dendritic locations and involve large glomerular synapses, similar to the types of synapses that relay information from the periphery to primary sensory thalamic nuclei. … (Levitt et al., 1995)

Here are a few parts that discuss this connection, but they don’t say whether it’s proximal.

Page 3:

CT cells that project to the posterior thalamic group (Pom), which does not receive primary sensory input, as well as to the VPm are found in deep layer 6. … (Zhang and Deschênes, 1997)

Page 3:

…while the shorter, lower layer 6 CT cells innervate interlaminar nuclei and/or regions of association thalamus affiliated with the primary sensory modality, in addition to providing reciprocal innervation to primary sensory thalamic nuclei (Deschênes et al., 1998; Llano and Sherman, 2008, see also Figure 4)

End of Page 9:

Similarly, in the somatosensory system, primary sensory cortical layer 5 projections to the secondary thalamic posterior group, Po, form clusters of large boutons, which are confined to the dorsal part of the nucleus. This part of Po, also receives input from layer 6 of the primary and second somatosensory areas and from the motor and insular cortices (Veinante et al., 2000).

Here are a couple of times it’s discussed in a summary without additional citation.

Summary on top of Page 10:

Secondary or association regions of thalamus receive their strong proximal inputs from another subclass of CT cells that do not project solely to primary sensory thalamus and from layer 5 cells in primary sensory cortex.

Figure 4 caption:

In association thalamic areas, the larger, more proximal synapses are provided by layer 6 CT cells and layer 5 corticofugal cells in primary cortex

Glances at some referenced papers

Usrey and Fitzpatrick, 1996, describing the striate cortex (a.k.a. V1)

End of page 1211, PDF page 9:

Smaller injections of biocytin restricted to either Vla or VIb resulted in different patterns of labeled terminals in the thal- amus. Figure 11 shows three examples of the distribution of layer VIb axons in the thalamus. These examples include a small population of layer VIb axons and two individual layer VIb axons. In each case, layer VIb axons give rise to collaterals with terminal boutons in the pulvinar nucleus before entering the LGN.

End of page 1212, PDF page 10:

These results suggest that neurons in layer VIa and VIb differ in their pattern of projection to the LGN. Neurons in layer VIa terminate in the principal layers of the LGN, whereas those in layer VIb terminate primarily in layers 3 and 6. In addition, layer VIb neurons send collaterals to the pulvinar nucleus.

Levitt et al., 1995 is not open access so I haven’t checked it.

So I haven’t found primary research saying that the input is proximal, but I haven’t looked closely yet.

[mrcslws]

And here’s some followup about the L5 → L4 “up the hierarchy” connection

Mentions in this paper

This connection is shown in Figure 4, but I’m not sure if it’s described in the text.

Here’s one part that it might be referring to:

The major extra-areal input to layer 4 of the second visual area, V2, is provided by CC projections from V1, ie. these inputs occupy the territory that is occupied by thalamic afferents in V1 … (Levitt et al., 1995)

CC cells? I’m not sure if it’s treating the “corticofugal” cell in L5 as a CC cell.

Also, Page 3: (no reference)

Unlike CT cells, CC cells have long, horizontally oriented axons that remain confined to the deep layers. From somatosensory cortex, for example, these branches project to the second somatosensory or motor cortices, or to the corpus callosum.

If CC horizontal axons can go up the hierarchy, and if L5 cells qualify as “CC cells”, maybe this is what it’s talking about?

As before, Levitt et al., 1995 isn’t open access, so I haven’t checked it yet.

[Timo]

Seems like consciousness can be lost in many ways. Video https://youtu.be/LMrzdk_YnYY?t=716 is saying that damage to the certain part of the brain stem will put you into coma. Is coma the same as consciousness switch off? That brain stem part of the brain is so important because it is an information hub.

Another youtube video https://www.youtube.com/watch?v=k_ZTNmkIiBc. As far as I understood, she is saying that your current consciousness is what ever is active in your brain at the moment. So consciousness is changing all the time even if it feels constant.

Your link is saying “Our findings suggest that the left claustrum/anterior insula is an important part of a network that subserves consciousness”. This supports the idea of consciousness being network of everything going on in your brain right now.

I guess this was off-topic post. Consciousness is so interesting subject that I could not stop myself.

Thanks for the claustrum link.

[dmac]

I found an interesting fact about the Thalamus -> Layer 6 connections, and that is that according to (Hunnicutt 2014, A comprehensive thalamocortical projection map
at the mesoscopic level, doi:10.1038/nn.3780), many thalamic relay cells preferentially innervate either the upper or lower layers of the cortex. It appears that the thalamus sends different information to L4 and L6.

[jhawkins]

Thanks for posting this. I hadn’t seen this paper. For a long time I have suspected that there should be differences between the origins of thalamic cells that project to upper and lower layers of cortex. One person I have asked about this, Murray Sherman, said that there isn’t a difference, that the same thalamic relay neurons project to both (mostly to upper layers, with a little to lower layers). The Hunnicutt paper introduces other, more complicated, possibilities. Figure 7g shows that Po and VPM project predominantly to upper layers with a small projection to lower layers. This is consistent with what Sherman says and what has been reported for sensory regions in other animals.

But the figure shows that motor cortex gets a massive input to L5b, which I believe in the mouse are the thick tufted cells that are the actual motor neurons. They show multiple thalamic nuclei projecting to the same motor cortex. I don’t know what to make of this.

[Bitking]

@jhawkins ,

From my reading I suspect that the L4 projections serve to establish and maintain thalamocortical resonance and the deep layer projections serve as a supervisory circuit for the feedback direction pathways.

See references 5 and 8 in this post:

HTMs, Time Locking, and Neural Oscillation (Brain Waves)

The basic HTM theory describes region updates as synchronous, with the SDRs as a static presence during each cycle evaluation; updated whole throughout the region after each evaluation. It is implied in all discussions for predictive dendrite segments (SDRs) that there is a “before” and “after” in regards to global states for the local region; the processing of the SDR neurons in a region could be considered as the evolution of the current region state to the next state as a coherent whole. I d…

[Casey]

@jhawkins

For a long time I have suspected that there should be differences between the origins of thalamic cells that project to upper and lower layers of cortex.

@jhawkins

Figure 7g shows that Po and VPM project predominantly to upper layers with a small projection to lower layers.

There might be separation. For primary cortex, POm has two zones. The anterior zone receives sensory input and from L5 TT cells, on the same cells. The posterior one only receives from L5 TT cells. The anterior one projects more to L5 than L1, whereas the posterior one projects mainly to L1 over broader areas. Cells that project to both tend to be closer to the other zone.

Convergence of Cortical and Sensory Driver Inputs on Single Thalamocortical Cells Convergence of Cortical and Sensory Driver Inputs on Single Thalamocortical Cells - PMC

A Morphological Analysis of Thalamocortical Axon Fibers of Rat Posterior Thalamic Nuclei: A Single Neuron Tracing Study with Viral Vectors Morphological Analysis of Thalamocortical Axon Fibers of Rat Posterior Thalamic Nuclei: A Single Neuron Tracing Study with Viral Vectors | Cerebral Cortex | Oxford Academic